SDKs - Antara

Antara speaks the OpenAI and Anthropic wire formats, so their official SDKs work as they are. Set the API key and the base URL:

SDK	Base URL	Endpoints
OpenAI (any language)	`https://api.antarah.dev/v1`	Chat Completions, Responses, Models
Anthropic (any language)	`https://api.antarah.dev`	Messages, count tokens. The SDK adds `/v1/messages` itself

from openai import OpenAI

client = OpenAI(api_key="ak_...", base_url="https://api.antarah.dev/v1")

stream = client.chat.completions.create(
    model="a1",
    messages=[{"role": "user", "content": "Write a haiku about the desert."}],
    stream=True,
)
for chunk in stream:
    print(chunk.choices[0].delta.content or "", end="")

import OpenAI from "openai";

const client = new OpenAI({ apiKey: "ak_...", baseURL: "https://api.antarah.dev/v1" });

const response = await client.responses.create({ model: "a1", input: "Write a haiku about the desert." });
console.log(response.output_text);

from anthropic import Anthropic

client = Anthropic(api_key="ak_...", base_url="https://api.antarah.dev")

with client.messages.stream(
    model="a1",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Write a haiku about the desert."}],
) as stream:
    for text in stream.text_stream:
        print(text, end="")

import Anthropic from "@anthropic-ai/sdk";

const client = new Anthropic({ apiKey: "ak_...", baseURL: "https://api.antarah.dev" });

const message = await client.messages.create({
  model: "a1",
  max_tokens: 1024,
  messages: [{ role: "user", content: "Write a haiku about the desert." }],
});
const text = message.content.find((b) => b.type === "text");
console.log(text?.type === "text" ? text.text : "");

Other tools that let you set an OpenAI- or Anthropic-compatible base URL work the same way.

What the SDKs get from Antara

Behavior	Detail
Error classes	Inference errors use each API’s native error shape, so `openai.RateLimitError`, `anthropic.AuthenticationError`, and the rest are raised as usual. See Errors
Retries	Every error response carries `x-should-retry: true` or `false`, which both SDKs follow. Only `429`, `409 idempotency_in_progress`, and retryable `5xx` errors say `true`
Request IDs	Every response has `x-request-id` and `request-id` (the same `req_…` value). Include it when you contact support
Rate-limit headers	`x-ratelimit-*` headers in OpenAI’s format, plus IETF `RateLimit` headers. See Rate limits

Fields that differ

Antara accepts each API’s official request schema. A few fields are handled differently; the API reference lists all of them per endpoint.

Endpoint	Not supported (`400 unsupported_parameter`)
Chat Completions	`n` other than 1, `audio`, `modalities` other than `["text"]`, `prediction`, `web_search_options`, `functions`, `function_call`
Responses	`previous_response_id`, `conversation`, `background: true`, built-in tools (`web_search`, `web_search_preview`, `file_search`, `code_interpreter`, `computer_use_preview`, `image_generation`, `mcp`, `local_shell`). Responses is stateless: `store` is always `false`
Messages	`top_k`, server tools (web search, web fetch, code execution), `document` / PDF blocks, `container`, `mcp_servers`

Fields Antara accepts but doesn’t forward, so SDK defaults never break a request: user, safety_identifier, store, metadata, service_tier (Chat Completions and Responses), and metadata, service_tier, cache_control, anthropic-beta (Messages). Prompt caching is automatic.

/v1/responses works only for models whose runtime serves the Responses API; others answer 400 unsupported_parameter with param: "model". Chat Completions and Messages work for every model.